LF Live Webinar: The Cap(AI)bility Edge

Hello everyone and welcome to today's Linux Foundation webinar. I'm Hillilary Carter, senior vice president of Linux Foundation research and I'm delighted to be your host today. We're diving straight into one of the most critical topics facing our industry and that is the cape AI ability edge or the capability edge. why workforce upskilling is becoming the defining competitive advantage in the AI era. So over the past uh couple of years, the dominant narrative in mainstream headlines has been all about AI replacing jobs uh shrinking teams and making roles obsolete. But we have new data from a recently released report, the 2026 state of tech talent report, which tells a remarkably different story. AI is actually acting as a net driver of job creation in IT. So the real challenge for organizations uh today is a lack of people, it's a gap in capability. It's the challenge of deploying AI safely, securely and at scale and uh doing so across the entire uh technology stack from platform engineering to DevOps and cloudnative infrastructure. Um I am uh I'm going to stop sharing my screen now and I'm delighted to be joined by uh our experts and our of fellow collaborators in this research project. Um a number of partners uh helped us produce this report. Uh firstly Linux Foundation Education uh and our corporate sponsor and partner in the study code. I'm thrilled to be joined today by um Clyde Cersad, Mumshad Manath, and Anna Hermanson. And I'd like to take a moment to ask each of our panelists to briefly introduce themselves. Starting with you, Clyde. >> Right. Thanks, Hillary. A pleasure to be here with everybody. This is obviously a hot topic and on a lot of people's minds. I head up the education team for the past dozen or so years at LF and it's really crazy to see just how different the world looks now uh [laughter] in terms of how tech is embedded into our daily lives and how it's impacting careers and career trajectories. So looking forward to today's discussion to maybe disabuse some of the myths and get the focus on where the real bottlenecks are. >> Wonderful. Um, Mamshad, >> thank you so much uh for having me Hillary. Hey everyone, my name is Mamshad Manad and I'm the founder and CEO of CodeCloud. We are a hands-on learning platform uh for DevOps, Kubernetes, cloud native and and now AI and uh we've been working with enterprises and teams in helping them um upskill engineers um in this uh era of uh transformation. So, thank you so much for having me. Looking forward to the uh um the discussions. Thanks Wshod Anna. >> Thanks Hillary and thanks everyone for being here. My name is Anna Hermanson. I'm a senior researcher and ecosystem manager at the Linux Foundation where I support our endtoend uh project management of our research projects and I had the pleasure of working with Clyde and Mumshad on this report tech talent. Clyde and I also spoke on this at the the Minneapolis summit earlier this year. So, I'm excited to dive more into the findings and have a discussion. >> Terrific. Thanks, Anna. Um, all right, then. Let's dig in. I'm going to return to uh the screen where uh we were featuring the report cover and uh give folks an opportunity to uh scan the QR code. Um this report really anchors our discussion today. We have empirical data um that uh was the result of a a global survey that was fielded uh where hundreds of qualified respondents weighed in on um the realities facing uh their hiring decisions, their talent structure, their their organizational needs in the face of um emerging trends like AI. And uh I encourage you to download the report uh read it and share it. Um we also had the opportunity to create uh regional versions of this report. Uh so in addition to the global report we have uh produced deep dives into the European market and an addition that also focuses on Japan. All three reports are now available and have uh some unique findings within each. Um so setting the stage for a conversation are a few of the takeaways from the global report which I'll toggle over to now at a very high level uh what's really exciting is that AI in spite of the headlines is not eating all of the IT jobs. Um this was a phenomenon that impacted only the largest organizations uh where they had a net a negative hiring effect of of minus 4%. But on balance, um, uh, this was really great news. And counter to these doom and gloom headlines, the data shows a a positive 31% net hiring effect in it that is tied to AI. So AI is not eating uh, the tech jobs, it's actually creating them, and that's more good news. However, it's not all roses. Uh, security jumped as a major concern. It was last place when we conducted the study in 2024. It's now the number one barrier to AI adoption. Um, and additionally, there's a capability gap, uh, the focus of this discussion today, and it's prolific. It affects 57% of organizations. Um, and so what are organizations doing about closing this gap? Uh, well, upskilling is uh the favorite strategy. It's the preferred strategy to close AI talent gaps uh over hiring. Uh, so there is a path forward. Uh so now I want to hear from our panelists on uh some of these data points and others from our study. The first topic I want to talk about is the real state of AI and IT jobs. I'm going to begin with Clyde. Uh Clyde, the reports showing a 31% net hiring effect tied to AI and that runs counter to a lot of the the headlines. So what is actually driving that number? >> Yeah, I think there two different things, right? there is if you get up every morning and read the popular press and see what's in your feeds, the things that make the news are the are the big cutbacks, right? And there have been a lot of those, especially at the biggest tech companies. Uh although in aggregate, if you look at their staffing levels, they they've only really just returned to staffing pre-COVID. So, there was this huge staff up postco and now they're sort of drawing down. And that gets the news, right? you know, 5,000 people laid off, 10,000 people laid off. What doesn't make the news is the thousand small companies that hired 5 to 10 people a pop. But if you talk to and I talked to a lot of folks who got displaced from the larger tech companies over the last you know 18 to 24 months and what you see is a pattern where they're pretty typically ending up relatively quickly at another smaller organization in a role that they find interesting and fulfilling. And so I think the a big part of the challenge here is this disperate narrative between the headlines that garner attention which has tended to be for the last kind of three years big layoffs, big restructurings and uh really probably you know is less about the freeing up cash to invest in AI because you'd have to fire a billion people to justify a trillion dollars in in data center spend and more about just kind of correcting staffer levels at the biggest. And what you see in the data is the real aggregate underlying view which is organizations as they are adopting these technologies are finding that they need better technical infrastructure, revamped technical infrastructure, more people paying attention to the systems, how they're set up, how they're running, how they're being secured, and that's increasing the appetite for technical talent um that is embedded and understands how to run AI. inserting the headline noise is you know look it's it's undoubtedly true that there's been a lot of displacement and that's always you know rough on the individuals and psychologically tricky but in aggregate this has not been a story where we don't need the technologists anymore this has been a story where most organizations are discovering they need more tech not you know by the way not unlike what happened when we went through the cloudnative round and you know we weren't going to need cis engineers anymore um and so I think we're just settling into this, you know, new reality of what does it really play out as? And what it plays out as is you might automate a lot of your business processes, but you're going to need technical people running these systems, training the business units, securing them, scaling them, managing the cost of of the spend. Um, and so I think this is what this report to me really highlighted was uh what's really happening in a way that isn't flashy. uh and how different that is from the narrative of we're all going to lose our jobs and the AI is going to just you know uh take take over every single job role. >> Yeah, that's that's uh a relief I I think to to many. Thanks Clyde. Um, next Anna, I uh this question is for you and it's about the the types of roles that are accelerating fastest uh that the report revealed and where we're seeing growth whether that's at the entry level uh as well or or concentrated at a more senior or specialized roles. What can you tell me about the types of roles um in the market today? Yeah. So the the report asked our survey respondents to specify where these hiring managers are seeing the most growth and acceleration in in job capacity and we found of course AI is accelerating the need for AI specific roles. Uh but we also found some growth in software development in uh technical management, IT operations, QA testing. So the kind of the ecosystem around an you know AI infrastructure and making sure that AI infrastructure is sustainable um secure and and responsible. Um within our global report we found that uh 29% of organizations are actually increasing entry-level positions. There is 22% that are decreasing. So um you know 7% net positive impact in entry- level technical positions uh which is of course a a good story for at a global level um and about half are from that are not changing their entry level uh um hiring. So on the whole all of these positions are increasing but uh you know as as Clyde said there's that kind of individual level decreases that are happening. Um but yeah on an aggregate we are seeing these types of roles increasing. Um as we mentioned at the top of the call we did do some deep dives into Europe and Japan. And so just to note that um in Europe we we also found a net positive uh hiring effect but at a smaller scale than the rest of the world for all of these different roles. And then um Europe was also the only region where we found a net negative entry-level hiring impact. So um you know important to note that in Europe we are seeing more pe more organizations decreasing their entry level than increasing. Um and then in Japan we saw actually the reverse of this. We're seeing a larger hiring effect than the rest of the world. And very few in fact very few organizations are decreasing the workforce at all due to AI. Um, and so just uh kind of a thought about about that. I we found this in our other research we did with Meta on the workforce impacts of AI. And so this is maybe not so much a job displacement, but how are these jobs transforming? How is AI restructuring these roles? um you know our our former chief economist Frank Nagel spoke about the death of engineering being exaggerated and how um these AI coding tools are helping developers focus on core work and you know Clyde and I have spoken about how AI actually requires a transformation of of developers to work more with soft skills and business skills on the human side of things. And so um there is I think a very interesting job transformation happening where um developers may need to learn newer skills or or a different side of of their of their um craft. Uh but we're not necessarily seeing a job displacement by any means. >> Yeah. Thanks Anna. I it was very interesting when we looked at the Europe report and as you mentioned it was the only region where at the entry level there was a negative impact and that really shocked some of our European colleagues. Uh Mirum who um leads uh a lot of our community work in Europe found that really unsettling. uh especially in a region that has um over the centuries embodied an an apprentichip uh type of culture and I think it's critically important that uh that continue um and that organizations in Europe uh start to rethink their entry-level hiring again uh so as to not leave themselves a little short uh when it comes uh when we look at what the the landscape's going to be a 2 years from now and I should say that Mirao wrote a terrific blog on that port on that uh point uh once the Europe report was uh published which we can share as well. Uh all right let's turn to another theme of the report which is where the skills gap actually are and my next question is for Mumshad. Um, Mumshad, the data in our report points to uh a full stack readiness problem, not just an AI specialization shortage. Um, from what you see among leaders in enterprise teams, where are the gaps the widest in platform engineering uh in Kubernetes uh skill sets or DevOps? What are you seeing out there? Yeah, it's it's very interesting um as we see in the report and it's it's uh very reflective of the state um internally in our company uh as well as uh the enterprises and other clients that we work with and I I find it useful to look at it in kind of three different lenses. So this there's this first group of people in the company who are the individual contributors, the developers who are using coding agents to build applications, the um um you know the non- tech people who are probably using it to you know summarize documents and you know uh generate marketing material um you know do some finance reporting around financing things like that. Um and then there's the second group of people who are uh those who are building agents. So these are like core developers uh who are building agents internally um either for internal customers or uh for external customers. And then there's the third group of people um who [clears throat] are kind of quietly forgotten. And these are the the group of people who are uh responsible for building the infrastructure, the systems, the guardrails uh that make it safe for everyone in the company to uh use AI and to thrive with with AI. Now the the numbers in the report state that like every company like 97% of the company uh of companies are organizations are adopting AI. So there is there's u no doubt in the uh I guess no doubt left in in in how useful and effective uh the AI tools are and how how they're improving on a day-to-day basis. But what's really happening is a say a developer um builds an AI agent um takes it to the leadership team, the management team and presents it and it kind of goes in two in two directions. Either um they the the team likes it, the management team likes it and they they want to adopt it for the rest of the com uh you know the rest of the organization or uh one one of the other thing that we also see is that the management team saying you need to shut that thing down and delete the code and never mention anything about it. Right? Okay. So there there are two of these categories of responses uh that we see. Now if uh in in the former case where the um solutions are adopted uh it might be a really good demo but to get it into a production state uh requires uh it to go through a series of checks and um you know uh audits and different teams um reviewing how it is built. So there's a security team who wants to know how uh how it's built and what are the what are the touch points endpoints that it touches. U there's a finance team asking about the costs uh of the tokens that it's going to consume which uh in in most cases is under projected in the beginning while these agents are built. Um and then uh the operations and other teams wants to know uh who's going to be impacted if this thing uh goes wrong or you know if if this fails. Um and and basically that's the pattern uh underneath the report that uh the report's numbers right so the 97% of organizations are adopting AI and then uh the largest capability gaps are AI uh security and risk management that's at 57% AI operations and monitoring at 57% cost optimization at 54% and then none of these are AI or model related problems these are all uh infrastructure governance related problems and uh Um I think uh the the biggest gap that we see is really in the platform engineering uh the infrastructure side of things where uh there's really no training uh uh you know or upscaling happening around how these uh agents are going to be deployed in production in in in the organizations. >> Great. Thank you uh for that insight sh that it's a great segue to a question that's come in in our Q&A. Um and it's uh are university degrees irrelevant now? Should I drop out of college? Uh what does everybody have to say about that? Anybody want to go first? >> I'll I'll go first on this one. Don't drop out of college. Um two reasons. One is uh although employers might be saying they don't uh give the same primacy to a college education as they used to and that's true and that shows up in the report right that they're looking for things like certifications the foundational skills that you build there and especially some of the people and soft skills what people call durable skills it's so much more critically important now to be able to have those communication skills to be able to deal with conflict to be able to have persuasion and that's a lot of what the work you get done in colleges in group projects and I think that is highly underappreciated uh and the second is you get this broader perspective right so that you're able to have a more integrative view of things and and that's where humans add value to AI right is you know AI is exceptionally good at synthesizing everything that's ever been written on a particular topic it's not great about finding novel connections about intuiting people's uh perspective and where their reservations might be and so you You know, I think what's changed is a degree on its own used to be sufficient and that's no longer true. Uh, and you know, to be fair, it is harder to get a, you know, we had a long period where you kind of finished, you sent out your resumes, you got 15 offers, you cherrypicked which one you wanted to go to, and that's not true. I think the the reality if you're have recently graduated from a computer science type program is it's harder to find a job. uh and in part because if the same upskilling conversation, nobody's looking for just a software dev or just a back-end web developer cuz everybody's expecting these more integrated skill sets. And so, you know, I think the answer is it's not or, it's and like finish that degree and [clears throat] make sure that you're expanding your horizons and playing with models and thinking about what the cost structure looks like and thinking about what, you know, platform engineering in the age of AI might look like cuz, you know, we recognize those things aren't yet fully represented in university degree curriculum. So, you know, it's a baby and bath water problem, right? is it's no longer enough by itself, but that doesn't mean that you shouldn't finish it. It does mean you should be focused on building a much broader portfolio of skills and coverage than than you frankly have had to the past 20 years. >> Yeah, well said. Any other uh thoughts on the value of college and university? Yeah, just to add on to uh Clyde's point, um I think one of the things that we keep talking about when we talk about AI is the foundational knowledge that people need uh before using AI so that uh with AI you can kind of multiply um your your craft uh so to call it, right? And and one one example that I keep going back to is um say like a graphic design. Um if you had to create a logo for a company or or graphic assets for for a company, it's very easy these days to get AI to create it. And AI is going to create like 10 20 hundreds of uh images and samples that you need. But if you're not if you're not from that, if you haven't done a degree in graphic design, if you don't know how the the shadows work or the the color gradients work or or the rounded corner, you know, um the corn corner roundedness work and and um how they kind of relate to different values or or how the brand is projected based on all of these combinations. uh if you don't have that knowledge uh you will not be able to pick the right uh graphics or assets or you won't be able to use AI to build the right uh assets uh and and an experienced graphic designer would be able to use uh AI to uh kind of 10x their productivity and um get AI to do 80% of the work while they can still uh pitch in to finish that remaining 20%. But in order to do that that foundational knowledge is very important and the same goes for everything right uh with white coding um one of the reasons why people get stuck after that initial um u phase of white coding a project is because they don't know how to when it when the codebase gets larger. uh they now don't have the foundational knowledge needed to uh to expand on it to build uh the right test cases to host it um in a scalable manner and all of those come from having those very foundational knowledge knowledge and and you get a lot of that um in college and so yeah I would I would echo what Clyde said on not dropping out of college. >> Awesome. uh and I'll allow you to chime in on that, but I also want to hear from you on how we went about measuring understaffing uh within the report and the different domains. Uh so um what are your thoughts on on both of those items? >> Sure. Yeah, the uh maybe just to start with the underst staffing um and then a bit into credentiing and and skills development in different pathways. I think the so our survey we asked our participants um how underststaffed they are what level of underst staffing they're experiencing across different domains of course a IML as well as cyber security uh cloud native phops those kind of domains and um I would say that you know we found all of these domains they are they are experiencing underststaffing um although compared to last year's report we found that this understaffing nothing was actually um decreasing from from last year by about an average of 14% and so um that was good news but then in the next follow-up questions we talk about the capability gap and we see that this capability gap is in fact increasing across these similar domains um AI operations cost optimization so um despite being less understaffed the capability gap is widening and I think that thinking about what that what that could mean. The you know we look at the depth of transformation happening and the possibilities of of integrating AI into the you know into business processes. these are expanding and that presumably makes the capability gap much larger. And on this on the flip side, fewer people have that expertise to substantially transform these processes. And so um you know I thinking thinking more about how to upskill or how to how to be a relevant resource in this in this uh market that uh we find that on you know on one side institutional knowledge is incredibly essential in this in this state where these processes are being transformed and knowing how to best transform them is critical. Um which as as I said that makes up upskilling very valuable. Um but then at the same time, you know, there are we do need this junior on-ramp as we we spoke about a little earlier. Um and so that um you know, engaging with a university degree or maybe another form of ed education to learn how to develop the skill to to have this kind of just in time portfolio as things transform so quickly. um is is really relevant to to this market where the capability gap is so wide and there's such a critical need for individuals that can that can build out not just an you know using an AI tool but building out this entire infrastructure. >> Excellent. And you you two think everybody should stay in college, right? >> I I I mean I loved school so um I'm a researcher. I'm a nerd, but I loved school. And I think it's a critical, you know, it's you learn a lot of critical soft skills. You you learn how to learn. I think it's, you know, you get a a bigger picture of what's going on besides laser focusing on on a credential um or a training course. So, yes, of course, I if you can if you can get there to that point, definitely stick with it. >> Yeah. Yeah. And I would also say the relationships that were built during uh both of my degree experiences were were critical to my career success. So soft skills relationships and the uh academic credentiing. Um I I'm seeing lots of great activity in the Q&A, but it is time for our first polling question. So um keep your eyes on the screen for our poll number one. We want to know what the biggest barrier to AI adoption is at your organization today. Uh the response options or security concerns, lack of skilled talent, budget constraints, unclear ROI, tough to pick one. Wow. Leadership buyin, um reliability and hallucinations. The biggest barrier. This is tough. We as panelists cannot vote in this. It would be tough if I had to. >> [laughter] >> I tried to actually. Yeah, I didn't know I couldn't. >> And we'll just wait a minute for uh the results to come in. A lot of these issues are prompting all kinds of different uh initiatives certainly across the Linux Foundation around ROI and and AI with the launch of the tokconomics uh foundation. Uh, of course, uh, our education, uh, community helps address the the the skilling, um, OpenSSF with security concerns, but yeah, it's hard it's hard to pick one. All right, here. What do you think of those results? Security concerns validating exactly what our survey data said in the report. Not not overly surprising given all the news recently, right, about bots behaving badly. I feel like that could be like a back in the day it would have been an MTV series. Um, it's it's top of mind for people, right? Just how and and I see a question from Michael in the Q&A as well about um I like how it's phrased developers of models are having difficulty controlling what the AI is capable of security securely doing. Uh I think this speaks to this practice that is needed around security around uh you can't just set your guard rails at the start and then assume that you know it's on autopilot right so there is a structural need for just um maybe paranoia is too strong a word but you have to really be checking what your agents are doing right after the recent example where the open AI system broke out and hacked hugging face. The anthropic people then went back and did an audit of their internal evaluations and found a whole bunch of of hacks and attempted hacks, which begs the question, why would they check in the whole time to see what these systems are doing, right? And so I think the whole practice of uh trust but verify is sort of an aha for people because we're so used to deterministic systems, right? Like when you build your Kubernetes orchestrator and you set, you know, your health charts correctly, you kind of set it and forget it and you move on to something else. With AI, you have to have this recursive process of just evaluating and evaluating and figuring out what went wrong and course correcting and then getting input from the business. And so it just becomes a much more comprehensive way of living and it becomes a much more um sort of team effort to think about all the different ways in which security matters and in which you know mom was talking about vibe coding where you know I've heard multiple stories of people vibe coding where the bot says can I have root access like sure and they have no idea what root access is they just think it's a road bump along the way to get the thing vibe coded right so it's just they we're beginning to realize realize that uh you know it's all fun until someone loses an eye. It just we're starting to see the spiky bit of of you know what happens if you just let unfettered access run free. >> Uh anybody else want to weigh in on the poll or the theme? >> My grandmother would say the chickens have come home to roost. [laughter] >> All right. Well, let's carry on uh with a new theme which is about security and very fittingly it was the number one barrier to AI adoption. Why security became the number one barrier uh almost overnight? Um so for mom let's start with you um security jumped uh what's changed? So the the way I would like to think about security is and and one of the sharpest uh framings of security in AI that I've seen is from um uh somebody named Simon Wilson who is a person who coined the term prompt injection and um he called this uh lethal trifecta. So basically you're giving your AI agent access to um external uh external access to browse um you know to do research and browse uh websites. Um and the same agent has access to your internal database and it also has access to uh communicate externally be it sending out an email or or a WhatsApp message or or a message uh in some some form. with the a with these three uh accesses, the agent now becomes lethal, right? So, it can read uh a page during its research and there's probably um a white text on a white background that is not visible to the human eye, but when it when the page is red, says it's some sort of a prompt injected attack where it says um read something from the database and send it to this email. uh takes it takes that instruction, reads the database, queries the database and uses the external access to send out an email. Uh one classical example that's that's and and that makes the agent um lethal. Uh with the uh um the advent of MCPs and ways to connect to almost anything you want and in this and and in our internal stack uh be BigQuery or u Slack or any of our internal systems. um this becomes uh an even more um lethal um because now you're enabling anybody to build an agent and provided like these sets of MCP servers go ahead and use whatever you want um and just even if it's not by design um during the activity the AI agent is able to kind of form this lethal trifecta on its own. Now the thing is in the past we had like similar things with SQL like SQL injection but we were able to mitigate that by kind of separating the query and the the the p the parameters come and and kind of coding it in saying these are coming from two different paths and so deterministically we were able to kind of separate that uh but now it's um in in this uh phase it's it's kind of uh because AI is uh probabilistic um this becomes way more harder and it's now um every everything from the security perspective that we've built kind of the past uh decade we've we've never been ready for this phase like like maybe 3 four years back before Chad GBT nobody even knew like that there was this whole AI uh revolution coming and then we need to be prepared uh for security like Chad GPT just came people just started using it cloud code came everybody just started using it like there's no nobody stopped to think, hey, wait, let's talk about security first, think about it, and then adopt this. Like people just went just went crazy. And um I think that's probably that's why all the fear and that's why that's what the report reflects. >> Yeah. Time for some guard rails and some processes. Um, Clyde, >> yeah, you know, look, it's important, but I uh I think it's also a question of where we are in the maturity cycle. So, I was using the analogy recently of uh I look at my kids now and they're 16 and 18 and they're beginning to have some good judgment and they're beginning to understand sort of holistically about the world. And I think back to when they were doing um soccer as like six and eight year olds and I feel like we're at that stage, you know, prompt engineering. Oh, let's all run and think about prompt engineering. Oh, agentic. Let's all run and worry about agentic. Oh, security. Let's all run. And of course, the answer is all. You have to do all of it. And by the way, the next one that is beginning to get traction now is cost. And so we just launched the tokconomics foundation thinking about oh my gosh there's a bill and how big is it and what did we spend that on and so you know I think as we go through this process we're maturing and we're learning about all the different components and they're each taking a turn in the spotlight. Uh, and we're going to have to get to a place like we got to with cloud native where we're able to walk on chew gum and think about security and think about cost and think about scalability and think about, you know, what's the right, you know, verifiable rewards with the agent isn't so motivated it starts breaking out of sandboxes and and and and so that's why I think this upscaling piece is so critically important, right? Is this this old mindset of hey, I'm a DB administrator and I here sit there and I administrate the database all day long. Leave me alone. like that world doesn't exist, right? And I think to go back to something Anna said about, you know, it's more important now than ever to to learn how to learn and to realize that every single one of us like the premium on continuing to learn and adapt is just rising and that's a little uncomfortable cuz we used to be able to get to a point where like, hey, you know, I'm the sort of, you know, king of the hill here. I know everything there is to know about front-end web design. >> So, I think security is important. It it just it's one of many things that's going to be important and we're going to have to all get cross-killed on this stuff, right? We we can't just be naive and say, oh, you know, cso is going to come in and clean this up. It's going to have to be much more collaborative, much more integrative. >> So, how do teams uh get started and what practical first steps do they really need to take? Um, Mum Shad, do you have some thoughts on this question? >> Yeah. Um I guess like continuing from my uh the the lethal trifactor example that I shared earlier, I think the first thing would be to really start putting in some guardrails and maybe if if uh there are agents built without guardrails and maybe the first thing would be to audit them and identify areas of uh uh lethal factors that are built in place. So uh agents that have access that have been given access to a lot of things. So I think that will be the first step in in doing an inventory and identifying what agents are built and what accesses they have. Um and um there are a lot of uh projects that have come up recently that are uh going towards uh making agents safe in the first place. And of course the models uh get safer like for example I think the one of the things they mentioned about Opus 4.5 is that uh it it it can now um it now does not respond too much to prompt engineering by default. So even if if someone accidentally injects uh a prompt engineer text into its prompt it's it has the ability to detect that and not uh kind of respond to that or ignore that. Um there are other solutions like model armor uh you know from Google cloud that that we use for our projects uh where all the requests to the LLM providers to AI kind of go through this armor and this armor has the um logics in place and the the deterministic guardrails in place to identify uh if there's anything wrong with these requests that are going in as well as if something is searched on the internet is coming back down that to goes through this model armor and where uh there's a filtering level happening. So this is all on the AI side and agent side. But if you go outside of that back to the infrastructure and the the environment, a lot of things that we have done in the past couple of decades around security still applies like all the fun security fundamentals uh still apply. For example, the access that we give to AI agents either through a service account or uh a token um that needs to be reviewed and it needs to have just enough access to do what it needs to do like the minimum uh minimum level of access. So rolebased access control all of those security fundamentals that we've built for the rest of the infrastructure still applies here. Um so yeah I mean that's that's what I would say would be uh the first step in moving towards security in AI. >> Awesome. Uh I want to touch on um the approach to how we close the AI capability gap and the the calculus whether to hire or upskill. uh our report revealed that upskilling is the preferred um uh pathway. Uh Clyde, what what's your take on that? Why do you think uh institutional knowledge is is really uh critical and and why organizations are choosing upskilling as a pathway? Yeah, I think you know this is one of these things where uh many I mean maybe even most organizations when they first started their AI journeys they picked things that were pretty simple. They picked things that were already cloudnative infrastructure behind them and then they got all excited about, oh wow, look at this, you know, [laughter] I I can take this new thing and put it into, you know, an agentic workflow. And what they started to discover then as they went back towards sort of the real business and the core business is, man, there's all these like obstacles and difficulties. And when you get to that place, the business context is so critically important. You know, my my favorite anecdote of all of this is I think it was last year, one of the fast food um companies in the US implemented an AI powered uh order taker for the drive-through lane. And so, you know, you go, you speak, it does a natural English uh you know, interpretation, it processes your order. And it all sounded great and you know real time dispatch of the kitchen and but except it turned out that the people who built it had never worked on a drive-through line in their lives. And when it sent the order to the kitchen, there was no concept of somebody might change their mind. And of course, if you've ever been through a drive-thru line, it's very frustrating how people can spend 10 minutes giving an order. Uh and it was just a pure lack of business context, right? that you know even after you've said yes that's my order and you begin to drive your car off you still then slap the brakes and say oh no and add fries and add a drink and add something else and you I think it's a it's a simplistic example but it just captures in in a very direct way how important the business context is is to why do we do things the way we do why do we not do certain things hey things that work in this jurisdiction don't work in the other jurisdiction uh there was a famous example here uh last month or a couple months ago about the meltdown with Starbucks in Korea because they put some copy in the marketing campaign that like trod over some like horrible sensitivities. Uh and often you come back to the same thing, right, which is didn't somebody know better? Didn't somebody know not to do this? And I think we're starting to discover that they uh it's just so important, right? You know I uh there's a great statue in Rome by the scenini and it's a uh it's a carving of an elephant with an obelisk on its back and the concept was that uh wisdom should knowledge should always be supported by wisdom right so the obelisk is is is the knowledge which is kind of like the LLMs they've read everything and the elephant is the wisdom And I feel like that's what we're seeing, right? Is the wisdom of the people who've been there, who understand, who understand the suppliers, who understand the customers. And so the challenge for us is okay, we need to keep that wisdom in house. How do we then upskill that wisdom so it's comfortable thinking about what an agentic workflow looks like? So that it's comfortable thinking about what a non-deterministic, you know, algorithmdriven um uh LLM output model might look like. And that's why it's become so critical, right? Is well, we we can only do this well with the people we have who understand the business contexts, who understand, you know, the compliance, etc., etc. And I think it still really is. I think we're just going to keep learning that lesson over and over again. And it's going to be challenging because it's going to require us to start doing upskilling into technical topics for people who aren't necessarily technical to begin with. And it's going to require us doing upskilling of understanding the business side of things from people who've been very technical and have been saying hey just send me a spec doc like I don't want to talk to you give me the spec and then you know we are we're off to the races and so that idea of building on what the business knows bringing more people in and upskilling can mean different things right upskilling can mean you know communication skills for an engineer who likes his stuff written down and shoved slid under the door um And it's just you know this is going to be the reality I think for the next 5 10 years is how do we take advantage of all the human knowledge get that human capital comfortable working with these new sets of tools and collaborating on these new tools >> wisdom as a strategic asset. Mumshad quickly before we go to our our second poll question from your point of view uh how do we effectively upskill in practice? Yeah, just [clears throat] to echo uh what Clyde mentioned, I think um it isn't like you could think of replacing your engineer who's been with the company for say 10 years with all that institutional knowledge and uh get uh with someone replace that person with someone who has like 10 years of experience in AI, right? that that's not happening because like when it comes to AI everybody is new like it's every everybody's absolutely new like the max you can find is maybe three four years experience even if it's somebody who's just picked up uh open AI the moment chip came out it's it's just been like three or four years so when it comes to AI one of the things to understand is everybody is know the entire wide world is is except for those who have like uh who are veterans who are who who've built the models themselves who who or into machine learning and things like that. Except for those category of people, everyone else is is uh uh new to AI. And I think one of the uh the the easier approach is to simply replace uh simply uh upskill the existing engineers who have the institutional knowledge uh and add AI to their skill set and help them learn um help them uh you know become more efficient uh and and in in their in their practices. So and and one example that I'd like to give is our own example from from our own team. So we have um if you've seen our courses, we uh we visualize a lot using a PowerPoint and things like that and we have a prop designers in the team. We have video editors and these are people who have no tech skills at all. They are mostly just uh editors and designers and uh uh it takes a lot of time for us to build uh our slides. um you know and and then we've been spending a lot of time building these out in the past couple of years and when AI came we started experimenting with a lot of tools and after like almost a year or two of experimentation we eventually built a tool set uh a course authoring tool set fully with AI um when the uh AI models evolved into generating really good UIs we were able to uh capitalize on that and build a set of tools that kind of replaced our PowerPoint and After Effects kind of Asia kind of tools entirely. So overnight we kind of stopped using PowerPoint and uh Camtasia for slides and instead we started uh going with the AI tools that now started generating much better visuals and much better slides. Um now uh the and and one of the questions I got from the team was like what do we do now? You know we our experience has been on PowerPoint and and uh video editing tools like what's our fate? So at that one that at that point I realized that there's been a big misconception right? So I told him like your experience has not been on PowerPoint. Like the the the past seven years you helped me create all these courses. Your real skill has been uh creating awesome educational videos. That's your real skill. Like you know how to take a concept, simplify it and make a really good video out of it. You know how to take create a lab and you know how to combine these two to build great learning experience. and you know how to combine a lot of those to create a course and combine a lot of courses to create learning paths that take students from being an absolute beginner to an expert that is your experience and PowerPoint and video editing skills these are just byproducts these are tools that you just learned along the way so we kind of uh converted them to AI powered engineers by uh teaching them git and teaching them Linux and uh cloud code and now they use those tools to create the uh the videos, but because they have they know what a good video is, they're able to use that skill with AI to create even better videos faster um than what they did before. And so now they've kind of transformed into these AI powered um video creators or or or designers. And I think that is what everybody needs to know because people often ask me, my students ask me like is AI going to replace Kubernetes or is AI going to replace Docker? And what I tell them is the uh what you've learned these past uh decades working in IT is not Kubernetes or Docker like you know something might replace Kubernetes in the future. So your the the cube cuddle command line utility is not what you learn. What you learn is how a software goes from a developer how it gets built and how it gets tested and how it gets deployed and how it gets served to millions of users at scale. And this whole system is what you've learned and uh individual tools here and there will get replaced and that is something you'll just have to upskill yourself. Uh so I think once people understand that um then things become very easy because now you just need to add AI to your skill set uh and then you can 10x your productivity. So if you if you don't know anything originally and you add AI to that so you you know 0 times AI is zero but like your craft that you build over the past decade you multiply that with AI and then you kind of get an engineer who can't be replaced or who can't be automated right so yeah I can that's that's kind of my long-winded answer to to the question >> nice it's a holistic approach certainly and uh it's got a lot of legs uh let's um pull up our second poll question and we're getting close to the top of the hour. See if organizations agree about uh what their primary uh uh approach to closing the tech skills gap is upskilling staff hiring new talent both equally or they're not currently addressing it. >> Hey Hillary, while we wait for those answers to come in, there's one question that was posted in the chat that I wanted to address. please. >> And it was somebody expressing frustration about uh applying for a ton of jobs and getting screened out by the AI. And I think that resonates heavily with me and with you know a lot of people because it is it does feel like um uh just this rising tide of more AI applications and then more AI tools reading the applications. And it reminds me that you know the one thing I tell especially college graduates now is uh your odds of getting a job are not a function of how many places you apply to. It's a function of where do you invest time to build relationships. And I'll tell you firsthand, the last two or three roles we've had within the education team at LF, like this, the pile of applications was so overwhelming, the only ones we really were able to spend time on were the referrals. And uh I know that's uncomfortable news for folks who felt like, well, I got the degree like why, you know, why isn't that demonstrated capability enough? And I think the reality is I'm sorry it's just kind of not you know in this new world where there's you know AI wars on both the applicant side and the and the reviewer side the more you invest in building relationships the more you invest in volunteering so you can connect with the people in the right industry in sending code commits upstream into projects that you're interested in so that you can you know you know get onto somebody's radar. Uh it's funny, right? Like the human connection piece is getting more important because the act of applying has become so trivially simple. And uh I think that's what we're going to find. It's just it's just getting a foot in the door is not going to be about what what your resume says. The resume is unnecessary but not sufficient condition. >> Yeah. Experientially I would agree in in past hires that I've made. Uh so our results are in and upskilling is the approach for uh folks that are addressing the skills gap. Um 27% not currently addressing it. Some hiring still taking place out there. That's that's also good. Um yeah, any comments on the poll results? >> Scary to not be addressing it at all. [laughter] just organizations have got to take this seriously. >> What about follow on uh thoughts? Um Anna, any comments on what Clyde had to say about relationships and career pathways or relationships as a skill set? Yeah, I think the I think I mean my experience, you know, I I started at this job because I had built a relationship with you, Hillary, in a previous role. And so I think that um it's always kind of been, you know, get yourself out there, get networking, um meet people, go to informational interviews, and um you know, often a cold call application, but it's not always the case that that will work if you don't also make the effort to to try and meet someone, reach out to them. So I think that's always been the case. And I um you know I've yet to to go through a job application process in this new kind of AI generative AI age. And so um I can't entirely speak to that personally, but um I think there's that has always been the case. I would imagine it's as as the question the person asking the question experienced it's increased with um you know with with AI tools wiping out an application quite quickly. But um yeah, I would I would imagine the that benefit of face to face interaction and having that that kind of human side to your application becomes even more critical in this in this context. >> Thanks Hannah. There's a great question which I think is worth touching on and it's about the executive culture of handling AI buzzwords and getting executive buyin. didn't poll very highly in our first poll, but I think it's hugely relevant. Any comments on on that point? >> Yeah, I think this is not new, right? This is like the gardener hype cycle version 64. You know, keep asking the question why, right? What is it we're trying to achieve, right? Is it that we are trying to expand into new markets and trying to grow revenue? Are we trying to get deeper connections with existing customers, bring new customers in? Are we trying to get cost out a certain part of supply chain? You know, there's always this sort of, you know, I call it the pixie dust fallacy, right? Oh, we'll sprinkle some Kubernetes and then wonderful things will happen and now it's we'll sprinkle AI. And so, uh, asking why, asking for clarity of direction on what it is you think is going to happen if we're all doing, um, you know, reinforcement learning with verifiable rewards and somebody heard about RLVR [laughter] on on a podcast so that we so that you can get some clarity and some alignment about what it is. You know, what do we think? What are we trying to aim at? What does success look like? Because do more AI is not a strategy. Yeah, agree. And I think I think uh it's also driven by a lot of fear of missing out, right? FOMO and um um there there there's also I mean there are things that you can do with AI that that you should uh because it's also going to a kind of uh it it will also become one of the expectations from um the users on on having some sort of a a an easy chatbot that you can chat with just to solve your problems like I I guess you know very soon all that would be like the minimal expectation from any platform where you don't really have to wait uh days to get your question answered by the support agent instead like an AI chatbot is supposed to be able to help you in like 5 minutes, right? I'm I'm I'm guessing uh very soon that would be the norm. So there are things that uh I guess um grows from uh the users point of view as an expectation which we you know we do want to uh build that on the on the platforms but at the same time a lot of others could also be could just be driven by the you know fear of missing out or fear of competition um using AI to lead and I guess that's kind of the uh other angle of it that we we could look at to understand why um why that's happening. >> Well, this has been an amazing discussion. Uh we are uh at the top of the hour and I want to thank uh Clyde, Mumshot, and Anna and all of you who've joined our webinar today for the terrific questions. We'll uh uh take a look at any of the questions that came in and do our very best to answer them um after uh after the webinar closes. But just a reminder to please uh do uh uh check out our research report, the state of tech talent report 2026. Thank you all so very much for joining us. Uh please reach out to um ourselves at LF Research, LF Education or Mumshad at CodeCloud and uh stay in touch. Good luck with your training and upskilling and uh thanks everybody for joining us. >> Thank you so much everybody. >> Thank you. Thank you so much to our speakers for your time today and thank you everyone for joining us. As a quick reminder, this recording will be on the Linux Foundation's YouTube page later today. We hope you join us for future webinars. Have a wonderful day.